
1.4.1. Impact of Drought on the Leaf Area Index in Yunnan Province, China#
Production date: 15-09-2025
Produced by: Amaya Camila Trigoso Barrientos (VUB)
🌍 Use case: Utilize satellite derived data to study the effect of drought periods on vegetation#
❓ Quality assessment question(s)#
Does version 3 of the C3S LAI dataset (derived from SPOT satellite imagery) provide sufficient temporal and spatial completeness to capture the impact of the 2009–2010 drought on vegetation in Yunnan Province?
How does the choice of relevant quality flags affect product spatial and temporal completeness?
The Leaf Area Index (LAI) and Fraction of Absorbed Photosynthetically Active Radiation (fAPAR) dataset, provided by the Climate Data Store (C3S), offers 10-daily gridded observations from 1981 to the present. It includes effective LAI values derived from multiple satellite sensors across different product versions. Effective LAI is defined as half the total surface area of photosynthetically active plant elements per unit of horizontal ground area. It relates to true LAI through a canopy-dependent structure factor. As a key biophysical parameter, LAI is widely used to assess vegetation status, monitor ecosystem dynamics, and inform environmental and agricultural decision-making.
This quality assessment uses version 3 of the LAI product, selected because it covers the study period from 2007 to 2013. During this time, data were obtained from the System Pour l’Observation de la Terre (SPOT) satellite, Vegetation (VGT) sensor, with a spatial resolution of 1 km and a 10-day temporal resolution. The dataset is evaluated over Yunnan Province, China, with particular attention to the 2009-2010 drought.
The objectives of this assessment are: first, to assess whether the product’s spatial and temporal completeness is sufficient to capture drought-related changes in LAI, which serves as a proxy for vegetation health; and second, to evaluate how different choices of quality flags influence the analysis.
📢 Quality assessment statement#
These are the key outcomes of this assessment
Applying a relaxed quality filter (bitmask 0x9C1) significantly increases LAI data availability during the wet season in Yunnan, maintaining over 84% completeness. However, this comes at the cost of increased uncertainty, as it includes pixels affected by albedo-related issues.
The conservative filter (bitmask 0xFC1), which excludes all flagged data, severely limits completeness during wet months, particularly from June to August, and especially in densely vegetated areas such as evergreen and mixed forests. However, this filter can be applied during the dry season with high completeness, while maintaining higher standards of data reliability.
The standardized LAI anomaly in Yunnan’s Tropic of Cancer zone was predominantly negative in March 2010, with extensive negative values also observed in the adjacent months, indicating that this period represents the strongest drought-related impact on LAI. This pattern is mostly consistent with the observations reported by Zhang et al. (2024) [1].
📋 Methodology#
The LAI variable from version 3 of the C3S product was downloaded for the period 2007-2013 and clipped to Yunnan Province, China. Two different quality filters were applied to the data: bitmask 0xFC1 (hereafter referred to as the conservative filter) and bitmask 0xFC1 (hereafter referred to as the relaxed filter). For each month in the study period, the percentage of completeness (the fraction of pixels that are not missing) was calculated under both filters for comparison. A date exhibiting a large difference in completeness between filters was selected, and the corresponding LAI map was visually compared with a land cover map to identify whether specific land cover types or zones were disproportionately affected by missing data. Monthly mean LAI over the entire province was then calculated and plotted for each year to assess whether the 2009-2010 drought is evident as lower LAI relative to other years. This analysis was performed for both filters to allow comparison. Finally, standardized LAI anomalies were calculated for Yunnan’s Tropic of Cancer zone, and the results were compared with those reported by Zhang et al. (2024) [1].
The analysis and results are organised in the following steps, which are detailed in the sections below:
Download LAI data from the SPOT satellite (VGT sensor) for the period 2007-2013.
Apply two distinct quality-masking approaches to filter the data.
Plot the percentage of valid LAI values over time for each approach.
Map LAI across Yunnan to compare the results from the two quality-masking methods.
Analyze the outcomes in relation to the dataset Documentation.
Calculate the spatial and monthly mean LAI, and plot the values for each year over the months of the year. Compare the results for the relaxed and the conservative filters.
Calculate the spatial and monthly standard deviation of LAI, and plot the values for the year 2010. Compare the results for the relaxed and the conservative filters.
Map LAI anomalies in Yunnan’s Tropic of Cancer zone for the drought period from August 2009 to October 2010, relative to the reference years 2007-2008 and 2012-2013.
📈 Analysis and results#
1. Request and download data#
Import packages#
Set parameters#
Define requests#
Define functions to cache#
Download and transform#
In this step, the dataset was spatially subsetted using a shapefile of Yunnan, an inland province in southwestern China. The shapefile was accessed from the GADM Database of Global Administrative Areas (version 4.1) via the UC Davis GeoData repository [2].
To filter out unreliable data, two distinct quality-masking strategies were applied:
Conservative: Retains only the highest-quality retrievals by applying bitmask 0xFC1. This mask excludes data flagged with any of the following bits: 0, 6, 7, 8, 9, 10, and 11, corresponding to missing input data, untrusted retrievals, unusable observations, inconsistent inputs, high albedo uncertainty (both snow and non-snow), and unreliable uncertainty estimates.
Relaxed filter: Permits retrievals with high albedo uncertainty (considered still usable), using bitmask 0x9C1. This mask excludes a smaller subset of flags: bits 0, 6, 7, 8, and 11, while allowing data flagged only for high albedo uncertainty in non-snow and snow conditions (bits 9 and 10).
These bitmasks are defined in Table 2, with flag definitions detailed in Table 7 of the Product User Guide and Specification PUGS.
100%|██████████| 84/84 [00:25<00:00, 3.27it/s]
100%|██████████| 84/84 [00:19<00:00, 4.34it/s]
2. Completeness analysis#
Filter data#
For each time step, the percentage of valid LAI data points within the Yunnan region is computed, and the results are plotted over time. This operation is repeated using both the conservative and relaxed filters.
The goal of this comparison is to assess how different treatments of uncertainty flags affect data completeness over Yunnan, particularly during periods of increased retrieval uncertainty such as the rainy season.
Figure 1. Percentage of valid LAI values within the Yunnan region using the conservative mask (0xFC1) and the relaxed mask (0x9C1). Percentages are shown over time, with points color-coded by month. Note that the y-axis scale differs between the two panels.
Using the conservative filter, the percentage of valid LAI data in Yunnan drops significantly between June and August, reaching levels below 20%. Adjacent months such as May and September also show reduced data completeness (see Figure 1, top pannel). This pattern corresponds to Yunnan’s marked wet and dry seasons [3], where the rainy season leads to a higher frequency of missing observations and albedo-related uncertainties, mainly due to cloud cover and wet canopy conditions.
In contrast, when applying the relaxed filter, the percentage of valid LAI data remains consistently above 84% throughout the period studied (2007–2013) (see Figure 1, bottom panel). This improvement is due to the relaxed mask allowing pixels flagged only for high albedo uncertainty in both non-snow and snow conditions (bits 9 and 10). In this context, bit 9 often relates to reflectance disturbances caused by clouds or wet vegetation, which are prevalent in Yunnan during the rainy season. As shown in Figure 15 of the PQAR, the study area is frequently affected by the activation of bit 9, especially during wet months. It can also be observed that the dry months at the end of the year (November–December) exhibit higher completeness for the relaxed filter, but the difference between wet and dry seasons is significantly less pronounced than under the conservative filter.
It is also instructive to investigate the spatial pattern of missing LAI values. Figure 2 shows LAI maps for August 10 from 2007 to 2013, comparing the conservative and relaxed filters. August 10 was chosen to exemplify differences in data completeness depending on the filter used beacuse it falls within the wet season and has low completeness under the conservative filter, as shown in Figure 1. The results highlight a strong contrast between the two approaches: the conservative filter produces highly incomplete maps across much of Yunnan in all years, whereas the relaxed filter provides nearly complete spatial coverage.
In some years, the conservative filter shows higher completeness in eastern Yunnan, where cropland, woody savannas, and grasslands are more prevalent than elsewhere in the province. Missing data under the conservative filter occur throughout Yunnan, particularly in high-LAI regions such as the mixed and evergreen forests in the west and southwest (see Figure 3).